In this application for a supplement to R37MH057881, we propose to extend our methodological work to make it relevant to Alzheimer disease (AD). For example, we have recently developed a method, called MIND, to estimate subject- and cell-type-specific (CTS) gene expression from tissue using an empirical Bayes method that borrows information across multiple measurements of the same tissue per subject. Using MIND to analyze multiple brain regions from the Genotype-Tissue Expression project (GTEx) reveals a subset of expression quantitative trait loci (eQTL) specific to certain cell types and others active across all cell types. Because the estimates are individual and CTS, they also reveal how genes are expressed and co-expressed over time; i.e., we can evaluate CTS expression and coexpression patterns in the aging brain, as well as relate them to genetic variation in the broad sense and to AD risk variation more specifically. Given the burgeoning data sets involving adult brain transcription, we envision the results from MIND to be a resource for the AD community. As currently envisioned, MIND requires three or more measurements of brain tissue from the same subject to produce good estimates of individual and CTS gene expression. This is a bottleneck for efficiently using transcriptomic data: many cohorts, such as CommonMind and the NIMH HBCC transcriptomic data sets, have only two regions assessed, namely the dorsolateral prefrontal cortex and the anterior cingulate cortex. We will extend the model underlying MIND to estimate CTS expression efficiently from two measurement of the transcriptome by using prior information within a hierarchical model framework. Our recent work supported by R37MH057881 has developed a method, called sLED, to compare covariance matrices and determine whether and how they differ. On a practical level, sLED has direct utility for determining whether gene coexpression matrices are distinct, such as from cases versus controls, and what genes drive these differences. Clearly sLED has direct applications to AD, not just for comparing coexpression patterns in cases versus control tissue and CTS expression, but also for more subtle genetic differences, such as how these patterns differ between carriers versus non-carriers of the APOE e4 haplotype. Combining results from MIND with analyses by sLED move these comparisons from the tissue level to CTS level. Additionally, we recently developed a method, called PisCES, to identify communities of closely interacting nodes within dynamic networks. The obvious application for AD is the detection of communities of highly coregulated genes within specific cell types as the brain ages; i.e., the dynamic network is the CTS gene coexpression network, and with particular interest in communities enriched for genes implicated in risk for AD. These results will yield important resources for the AD community regarding the dynamics of the aging brain and how it relates to genetic risk for AD.